I'm reminded of the 'cstross quip, "Libertarianism is like Leninism: a fascinating, internally consistent political theory with some good underlying points that, regrettably, makes prescriptions about how to run human society that can only work if we replace real messy human beings with frictionless spherical humanoids of uniform density."
Right. In cases like this, you either a) have unmodeled dynamics or b) (much less likely) you found a superior strategy.
If b), try it out in the wild. If a), add assumptions that make the observed behavior optimal again, and see if those assumptions apply to the real world situation.
EDIT: corrected the quote.